Android Malware Detection Using Category-Based Permission Vectors

With the drastic increase of smartphone adoption, malware attacks on smartphones have emerged as serious privacy and security threat. Kaspersky Labs detected and intercepted a total of 5,730,916 malicious installation packages in 2017. To curb this problem, researchers and various security laboratories have developed numerous malware analysis models. In Android based smartphones, permissions have been an inherent part of such models. Permission request patterns can be used to detect behavior of different applications. As applications with similar functionalities should use permission requests in similar ways, they can be used to distinguish different types of apps. However, when analysis models are trained on permission vectors extracted from a mixture of applications without maintaining any differences that naturally exist among different application categories, aggregated results can miss details and this can result in errors. In this paper, we propose a permission analysis model for android applications which includes a classification module and a malware detection module based on application permission vectors to deal with Android malware detection problem. We mine the benign application permission vector set into 32 categories by mining the similarity of permission vectors, and input malicious application permission vector sets into the model to obtain class labels, then extract sensitive features from different classes. Finally, sensitive features of each class are respectively input into the machine learning algorithm to obtain a classification model of malicious and benign applications. Our experimental results show that our model can achieve 93.66% accuracy of detecting malware instances.